Ana Hobden

Hmm, so `ConfChange` should normally come from Raft log entries. This is the replicated log, so 'injecting' an entry is inherently unsafe. I suggest you remove that code. (https://github.com/wayfair-incubator/uring/blob/a0a5ffa2990e2e8c91f9849cd7eefdd082501543/src/main.rs#L557-L584) :)...

Also: I'm sorry the docs aren't so clear. :( We are trying to improve it and would definitely love any concrete issues/prs with improvements.

You should definitely be able to grow/shrink your Raft cluster online. :)

Hmmm, that's a bit of a puzzle isn''t it! I see that index 5 is done: ```bash Nov 07 12:17:02.582 INFO [commit: 5, term: 3] restored snapshot [index: 5, term:...

Yup I misread and deleted the comment :-P

@Licenser If you don't call `curl -X POST 127.0.0.1:8081/node/1` in your reproduction case it seems to work without panicking... At least raft is *trying* to send the correct messages but...

Yeah, let's get this working! :) Would love to see more raft-rs derivatives and this is informing how we can better document things. (Un)Fortunately we're updating our membership change mechanisms...

> bootstrap 1, join 2 1, join 3 1, kill 1, join 1 2 ends up in "The node 1 already exists in the voters set." At the `join 1`...

> This is where it gets problematic, after recovery the nodes either run into a "node already known" problem when the ConfState is persisted to disk and re-loaded or it...

One problem with isolating these sorts of details to `#[cfg(test)]` only is you cannot use them in benchmarking code. In order to refine this we'll probably need to have some...